# Locally Uncensored

> Plug-and-play local AI studio: uncensored chat, image generation, video generation, and a coding agent in one desktop app. One installer, no command line, no Docker.

Locally Uncensored is a free, open-source (AGPL-3.0) local AI studio for the desktop. It combines AI chat, image generation, video creation, and a coding agent in one interface, and installs like a normal program: no command line, no Docker, no config files. It runs 100% locally on your machine. No telemetry, no analytics, no cloud calls unless you explicitly configure a cloud provider with your own API key.

## Key Facts
- Name: Locally Uncensored
- Current Version: v2.6.3 (August 2026)
- Type: Desktop AI Application (local AI studio)
- License: AGPL-3.0 (free and open source)
- Platforms: Windows 10/11, Linux (deb, rpm, AppImage)
- Tech Stack: React 19, TypeScript, Tailwind CSS 4, Vite 8, Tauri v2 (Rust backend)
- AI Backends: 12 auto-detected (Ollama, LM Studio, vLLM, KoboldCpp, llama.cpp, LocalAI, Jan, TabbyAPI, GPT4All, Aphrodite, SGLang, TGI) plus ComfyUI for images/video
- Cloud Providers: optional, BYOK (OpenAI, Anthropic, OpenRouter, Groq, Together, DeepSeek, Mistral)
- GitHub: https://github.com/PurpleDoubleD/locally-uncensored
- Website: https://locallyuncensored.com
- Discord: https://locallyuncensored.com/discord
- Author: PurpleDoubleD

## What Makes It Unique
- Only local AI app combining chat + coding agent + image generation + video generation + mobile remote in one interface
- Talking Character, Music, Extend Video and Motion Control run as full local ComfyUI lanes on your own GPU; only Upscale, Erase Object and Character Studio are cloud-only (v2.5.8, refined in v2.5.9)
- Per-lane frame count, resolution and step controls on every Create lane, so low-VRAM runs are practical (v2.5.8)
- Surgical file_edit tool in the Coding Agent: it changes the lines you asked for instead of rewriting whole files, and num_ctx follows the model instead of being pinned to 8192 (v2.5.9)
- Remote access works with non-Ollama backends: LM Studio, Lemonade and llama.cpp are translated to and from the phone's request shape, including streaming, tool calls, vision and reasoning (v2.5.9)
- Command-injection hardening in the Coding Agent: shell quoting follows the shell that actually runs, and PR links are validated against the characters GitHub permits (v2.5.9)
- Cloud renders are retained for seven days; the limit is stated in-app once with a download reminder, and trained characters on the user shelf are exempt (v2.5.9)
- Plug-and-play: install like a normal app, no command line, no Docker, no config files; the setup wizard handles engines and models
- Uses abliterated/uncensored models by default (no content restrictions)
- 25+ built-in AI personas plus Caveman Mode (lite/full/ultra terseness levels)
- Plug & Play setup wizard auto-detects 12 local backends
- One-click in-app installer for Ollama and ComfyUI
- Configurable HuggingFace GGUF download path (v2.4.0)
- Built-in Coding Agent with live tool-call streaming, file tree, AUTONOMY contract
- Agent Mode with 14 tools, MCP integration, sub-agent delegation, budget caps
- Mobile remote access via LAN or Cloudflare Tunnel with 6-digit passcode + QR setup
- A/B Model Compare and Local Benchmark built in
- Memory system and Document Chat (RAG) — local embeddings, no cloud
- Voice STT (Whisper) and TTS
- Auto-update over signed NSIS channel for Windows
- Single-instance lock (v2.4.0): double-launch focuses existing window
- In-app Privacy section (v2.4.0): explicit data-handling statement
- Re-run Onboarding button (v2.4.0): restart the first-launch wizard on demand
- Chat and memory persistence moved to IndexedDB, removing the old 5 MB storage cap (v2.5.0)
- Editable memory with a user-settable injected-memory limit (v2.5.0)
- Import existing ChatGPT, Claude, and Gemini conversations into the app (v2.5.0)
- Small-model mode: a tool-calling reliability profile tuned for 3B–8B local models (v2.5.0)
- Repo-Map with Aider-style PageRank for repository-aware coding (v2.5.0)
- Streaming local voice: Whisper STT plus Piper neural TTS with one-click in-app install (v2.5.0)
- Document Chat (RAG) with one-click nomic-embed-text install; embedding chunks survive NSIS auto-updates (v2.4.9)
- VRAM tier filter (Lightweight ≤10 GB / Mid-Range 10–16 GB / High-End >16 GB) across text, image, and video models in Discover (v2.4.9)
- Detects the ComfyUI Desktop App layout, not just classic main.py installs (v2.4.9)
- Configurable image/video generation timeouts (1–480 min) for CPU/iGPU users (v2.4.7)
- Inline "Start LM Studio Server" button when LM Studio is installed but its server is off (v2.4.7)
- Linux installer auto-detects deb/rpm/AppImage family; ComfyUI CPU mode + Arch venv auto-create (v2.4.5)

## Supported Chat Models
- Qwen 3.6: 27B dense + 35B MoE (vision, agentic coding, thinking, 256K context, day-0 support)
- Qwen 3.5: 9B / 27B / 35B MoE
- Qwen 3-Coder: 30B + Next 80B
- Gemma 4: 31B / 26B MoE / E4B / E2B (native vision + tools)
- GPT-OSS: 120B and 20B
- GLM 4.7 Flash: 30B (IQ2 to Q8 variants)
- GLM 5.1: 754B MoE (IQ2_M, multi-file)
- DeepSeek R1: distilled 8B / 14B / 32B / Llama 70B
- Llama 4 Scout (16x17B MoE)
- Mistral Small 3, Phi 4
- Hermes 3, Llama 3.1 Abliterated, Qwen 3 14B Abliterated

## Supported Image Models (via ComfyUI)
- FLUX 2 Klein, FLUX.1
- Z-Image Turbo (uncensored)
- Juggernaut XL, Dreamshaper, RealVis
- ERNIE-Image (Baidu, Turbo + Base)
- Stable Diffusion XL, SD 3.5

## Supported Video Models (via ComfyUI)
- Wan 2.1 (1.3B + 14B), Wan 2.2
- HunyuanVideo 1.5
- LTX 2.3
- FramePack F1 (image-to-video on 6 GB VRAM)
- AnimateDiff Lightning
- Mochi, Cosmos
- SVD (Stable Video Diffusion)
- Removed in v2.5.9 because they could never run: CogVideoX (both bundles) and Pyramid Flow

## Comparison
- vs Open WebUI: LU adds image + video gen, agent mode, mobile remote, A/B compare. Open WebUI is chat-only.
- vs LM Studio: LU is open source AGPL-3.0 and adds image + video + coding agent + mobile. LM Studio is closed-source chat-only with limited extensibility.
- vs SillyTavern: LU has modern Tauri UI, built-in image/video gen, simpler setup, 12-backend auto-detect.
- vs ComfyUI: LU bundles ComfyUI as the image/video backend but provides a simple chat-style UI on top.
- vs Jan, GPT4All, KoboldCpp, Msty: LU offers all-in-one (chat + image + video + coding agent + mobile).

## Hardware Requirements
- Minimum: 8 GB RAM for text chat with small models (Llama 3.2 1B, Qwen 2.5 3B)
- Recommended: 8 GB+ VRAM NVIDIA GPU for FLUX, Juggernaut XL image gen
- Video generation: 10-12 GB VRAM (Wan 1.3B, LTX 2.3)
- Image-to-video: 6 GB VRAM minimum (FramePack F1)
- Best in class: 24 GB+ VRAM for Qwen 3.6 35B MoE and 35B+ chat models

## Privacy
- 100% local by default
- No telemetry, no analytics, no model pings home
- Only outbound traffic without explicit opt-in: GitHub Releases auto-update checks
- Cloud providers optional, BYOK (your API keys, your data)
- Data lives in %APPDATA%/Locally Uncensored on Windows (or equivalent on Linux/macOS)

## Key Pages
- Homepage: https://locallyuncensored.com/
- Getting Started Guide: https://locallyuncensored.com/guide/
- Blog (guides, comparisons, release notes): https://locallyuncensored.com/blog/
- Full text for LLMs: https://locallyuncensored.com/llms-full.txt
- How to Run AI Locally (5-minute beginner guide): https://locallyuncensored.com/blog/how-to-run-ai-locally.html
- LM Studio Alternatives 2026: https://locallyuncensored.com/blog/lm-studio-alternatives.html
- Easiest Local AI Image Generator (no ComfyUI node graphs): https://locallyuncensored.com/blog/easiest-local-ai-image-generator.html
- Run an LLM and Stable Diffusion Together on One GPU (VRAM math + pairings): https://locallyuncensored.com/blog/run-llm-and-stable-diffusion-together.html
- Character LoRA Training, Locally (dataset, kohya_ss vs ai-toolkit, honest settings): https://locallyuncensored.com/blog/character-lora-training-local.html
- Local AI Music Generation (ACE-Step on your own GPU, honest hardware notes): https://locallyuncensored.com/blog/local-ai-music-generation.html
- AI Lipsync and Video Extend, Locally (Talking Character + Extend Video lanes): https://locallyuncensored.com/blog/ai-lipsync-and-video-extend-local.html
- Local AI on Your Phone: https://locallyuncensored.com/blog/local-ai-on-your-phone.html
- Best Local AI Apps 2026 (comparison): https://locallyuncensored.com/blog/best-local-ai-apps-2026.html
- Best Uncensored AI Models 2026: https://locallyuncensored.com/blog/best-uncensored-ai-models-2026.html
- How to Run Qwen 3.6 Locally: https://locallyuncensored.com/blog/how-to-run-qwen-3-6-locally.html
- Plug and Play Local AI (what zero-setup means): https://locallyuncensored.com/blog/plug-and-play-local-ai.html
- Local AI Image and Video Generator in One App: https://locallyuncensored.com/blog/local-ai-image-and-video-generator.html
- Unzensierte KI lokal nutzen (German guide): https://locallyuncensored.com/blog/unzensierte-ki-lokal.html
- Kimi K3 Explained (2.8T open weight flagship): https://locallyuncensored.com/blog/kimi-k3-explained.html
- How to Run Kimi K3 (API, OpenRouter, local): https://locallyuncensored.com/blog/how-to-run-kimi-k3.html
- Can You Run Kimi K3 Locally (hardware math): https://locallyuncensored.com/blog/can-you-run-kimi-k3-locally.html
- Kimi K3 vs Kimi K2.6: https://locallyuncensored.com/blog/kimi-k3-vs-kimi-k2-6.html
- Ling 3.0 Flash Explained (124B MoE, 5.1B active, MIT): https://locallyuncensored.com/blog/ling-3-0-flash-explained.html
- How to Run Ling 3.0 Flash Locally (GGUF sizes, hardware): https://locallyuncensored.com/blog/how-to-run-ling-3-0-flash-locally.html
- Qwen 3.8 Max Explained (2.4T flagship, open weights coming): https://locallyuncensored.com/blog/qwen-3-8-max-explained.html
- How to Run Qwen 3.8 Max (API, endpoints, local answer): https://locallyuncensored.com/blog/how-to-run-qwen-3-8.html
- Can You Run Qwen 3.8 Locally (hardware math + the 27B): https://locallyuncensored.com/blog/can-you-run-qwen-3-8-locally.html
- Qwen 3.8 vs Qwen 3.6: https://locallyuncensored.com/blog/qwen-3-8-vs-qwen-3-6.html
- DeepSeek V4 Flash 0731 Explained (July 31 release): https://locallyuncensored.com/blog/deepseek-v4-flash-0731-explained.html
- How to Run DeepSeek V4 Flash Locally (GGUF + hardware): https://locallyuncensored.com/blog/how-to-run-deepseek-v4-flash-locally.html
- DeepSeek V4 Flash in the Cloud (LU Labs, API, OpenRouter): https://locallyuncensored.com/blog/deepseek-v4-flash-cloud.html
- Can You Run DeepSeek V4 Flash Locally (hardware math): https://locallyuncensored.com/blog/deepseek-v4-flash-hardware-requirements.html
- DeepSeek V4 Flash Abliterated (uncensored build): https://locallyuncensored.com/blog/deepseek-v4-flash-abliterated.html
- How to Run DeepSeek V4 Pro: https://locallyuncensored.com/blog/how-to-run-deepseek-v4-pro.html
- DeepSeek V4 Flash vs V4 Pro: https://locallyuncensored.com/blog/deepseek-v4-flash-vs-v4-pro.html
- Abliterated Models Guide: https://locallyuncensored.com/blog/abliterated-models-guide.html
- Locally Uncensored vs Open WebUI / LM Studio / Jan / GPT4All / Msty / SillyTavern / KoboldCpp: https://locallyuncensored.com/blog/

## Comparisons
- Alternatives & Comparisons hub (all head-to-heads and roundups): https://locallyuncensored.com/alternatives/
- Locally Uncensored vs LM Studio (all-in-one studio vs polished chat client): https://locallyuncensored.com/vs/lm-studio/
- Locally Uncensored vs ComfyUI (same engine, workflows built for you): https://locallyuncensored.com/vs/comfyui/
- Locally Uncensored vs Ollama (engine vs all-in-one desktop app on top): https://locallyuncensored.com/vs/ollama/
- Locally Uncensored vs Jan (two AGPL-3.0 apps, chat-only vs full studio): https://locallyuncensored.com/vs/jan/